Leisure-Time Physical Activity May Attenuate the Impact of Diabetes on Cognitive Decline in Middle-Aged and Older Adults: Findings From the ELSA-Brasil Study

Dementia is a global public health emergency (1). In 2019, ∼55 million people were living with dementia worldwide, a number expected to triple by 2050, reaching >152 million (2). The increased prevalence of risk factors for dementia, including diabetes and physical inactivity, may be responsible for 50% of new cases in low- and middle-income countries (LMICs) (3). In Brazil, 5–10% of the total cases of dementia (4,5) and one in 10 cases of diabetes can be attributed to physical inactivity (6).

Physical activity may counteract the risks associated with sedentary behavior and reduce the overall risk of all-cause dementia (7). Physical activity attenuated the impact of cardiovascular risk factors (8) and aging (9) on the incidence of dementia in older adults. Similarly, findings from the UK Biobank involving 90,320 middle-aged and older adults revealed that in sedentary individuals, those with low physical activity had an increased risk of all-cause dementia and Alzheimer disease, whereas participants with high physical activity levels did not (7).

Diabetes has been associated with an increased risk of dementia (4,5,10). Various neuropathological mechanisms might explain this association (10). Disruption in cellular metabolism caused by altered insulin action and diabetes leads to high production of β-amyloid protein and impaired cerebral blood flow, two mechanisms believed to be central in cognitive impairment and dementia (10). Small clinical trials in older adults (11,12) and one longitudinal follow-up of 189 adults with diabetes (13) have shown the interaction of the simultaneous presence of multiple risk factors for dementia in individuals with diabetes. However, there has been little investigation of this in a free-living population setting. Furthermore, no such studies have been conducted in cognitively diverse populations with and without diabetes in LMICs, where the main projected burden of dementia lies (14). Also, whether (and to which extent) leisure-time physical activity (LTPA) is associated with cognitive impairment in individuals with diabetes remains incompletely investigated and minimally characterized.

Therefore, this study aimed to investigate the joint association of diabetes and LTPA with cognitive decline in middle-aged and older free-living adults participating in the Estudo Longitudinal de Saúde do Adulto (ELSA-Brasil) study.

ELSA-Brasil is an ongoing multicenter cohort study investigating risk factors and determinants of chronic diseases in a free-living Brazilian population (15). Baseline data were collected from 15,105 active and retired employees (age 35–74 years) from public universities and research institutions in the capitals of six Brazilian states (São Paulo, Rio de Janeiro, Minas Gerais, Espírito Santo, Rio Grande do Sul, and Bahia). Trained and certified research assistants collected data via face-to-face interviews and clinical tests in clinical research centers (15). The study was approved by local research ethics committees (São Paulo, Rio de Janeiro, Minas Gerais, Espírito Santo, Rio Grande do Sul, and Bahia). Participants signed consent forms before data collection in each of the three waves.

For the current study, we analyzed data from baseline (2008–2010) and wave 3 (2017–2019). We included participants with complete information about physical activity (baseline), cognitive function (baseline and wave 3), and covariates (baseline).

Diabetes at baseline was defined as having one or more of the following criteria: HbA_1c ≥6.5%, fasting glycemia ≥126 mg/dL, 2-h value during a 75-g oral glucose tolerance test ≥200 mg/dL, use of glucose-lowering medication, or self-reported medical diagnosis of diabetes.

LTPA at baseline was assessed using the International Physical Activity Questionnaire (long form). The instrument was previously validated for the Portuguese-speaking population and had moderate to strong concordance with accelerometer-based measurements (16). Participants answered questions about the frequency and time spent per day walking and in moderate to vigorous physical activity during leisure time, permitting their characterization as active (≥150 min per week) or inactive (<150 min/week) based on the 2020 World Health Organization guidelines for physical activity and sedentary behavior.

We used six standardized cognitive function tests for the Brazilian population: learning, recall, and word recognition tests; semantic and phonemic verbal fluency tests; and the Trail-Making Test Part B (TMT-B). Tests were applied in a quiet room by trained interviewers using standardized protocols and were supervised using recordings for quality control (17). Memory was examined using a validated adapted version of the Consortium to Establish a Registry for Alzheimer’s Disease assessment. The memory test score represents the number of correct words in the three tasks examined (learning, recall, and word recognition). We made changes in the content of memory tests to minimize previous testing affecting later test results, for example, changing the words to be recalled. Semantic and phonemic verbal fluency tests evaluated language function. The semantic verbal fluency test asked participants to name as many animals as possible. The phonemic test asked them to say as many words starting with a specific letter as possible in 1 min. The TMT-B examined attention, concentration, psychomotor speed, visuomotor scanning, and mental flexibility. The score was the time (in seconds) spent completing the test. It was multiplied by −1 so that scores closer to 0 indicated better performance. Also, because of its skewed distribution, the score was log transformed. Participants completed a smaller TMT-B with the same dynamic to familiarize them with what was expected. Previous results from ELSA-Brasil showed that the learning, recall, and word recognition tests had moderate reliability, the verbal and semantic fluency tests had good reliability, and the TMT-B had almost perfect reliability (18).

Scores from each domain (memory, language, and TMT-B) were adjusted for age, sex, and education using the predict command in STATA. Next, we standardized the predicted value in z scores based on the sample mean and SD. Then, we averaged the domain-specific z scores to create a global score, which was also standardized, resulting in a scale with mean of 0 and SD of 1 (19). We opted to assess the diabetes/physical activity interaction in association with the global instead of domain-specific cognitive function score for two reason: 1) the lack of pathophysiological mechanisms supporting a differential association with physical activity and 2) the fact that scores vary in the same direction according to physical activity levels (20).

Lower scores indicated worse global cognition function. Incident cognitive impairment was defined as a global cognitive function score lower than −1 SD from baseline mean (21).

Covariates were also obtained in the clinical research centers via face-to-face interviews and clinical assessments. Sociodemographic characteristics included study center, age, sex, education, and self-declared race/skin color. Health behaviors included smoking and excessive alcohol consumption (≥210 g/week for men and ≥140 g/week for women). Alcohol consumption was assessed using a questionnaire including beverage type, usual dose, and frequency. Presence of common mental disorders was defined as a score ≥12 on the Clinical Interview Schedule–Revised. Weight and height were obtained while fasting, and BMI was calculated as kg/m². Blood pressure was measured using a validated oscillometric device (Omron HEM 705CPINT) on the right arm after a 5-min rest. Individuals with elevated systolic (≥140 mmHg) or diastolic blood pressure (≥90 mmHg) and those using antihypertensive medication were classified as having hypertension.

Analyses were conducted using STATA software (version 14.2). Categorical variables are expressed as frequencies and percentages. Continuous variables are described using mean and SD or median and interquartile range, depending on the normality of the data distribution, assessed using histograms and asymmetry coefficient skewness. Statistical significance was defined as α level of 5%. The significance level for interaction tests was P value <0.10.

We used linear mixed-effects models with robust SEs and random intercepts and slopes to investigate the relationship between LTPA, diabetes, and cognitive decline across follow-up. The time scale in this analysis was the participant’s age in each wave. The longitudinal joint association between LTPA and diabetes with cognitive decline was evaluated graphically by adding the interaction terms of LTPA and diabetes with the time scale. Models were adjusted for age, sex, education, race/skin color, study center, smoking, excessive alcohol consumption, obesity, hypertension, and common mental disorder.

Associations with incident cognitive impairment were modeled with Cox stratified proportional hazards models, considering participant time since baseline visit as the time scale and excluding participants with cognitive impairment at baseline. We assessed the proportional hazards assumption of our models by testing the correlation of scaled Schoenfeld residuals with time and through graphical evaluation of the Schoenfeld residuals. Covariates included in models were baseline age, sex, education, race/skin color, smoking, excessive alcohol consumption, BMI, common mental disorder, and hypertension. Also, we included an interaction term between LTPA and diabetes to evaluate possible effect modification. We chose to use Cox stratified models because of the apparent nonproportionality of the hazard for the study center variable. We used Laplace regression models to estimate the difference in years of the onset of cognitive impairment associated with LTPA and diabetes status. Percentiles of time to cognitive impairment were modeled as a function of physical activity and diabetes status while adjusting for potential confounders. We fitted Laplace regression models on the 50th percentile of age at cognitive impairment. Results were interpreted as the difference in years of age at the onset of cognitive impairment. Participants with cognitive impairment at baseline based on this cutoff were not included in the Cox or Laplace analyses.

We performed a series of sensitivity analyses to verify the consistency of our findings. First, we used previously validated educational level– and age-specific cutoffs for cognitive impairment for the ELSA-Brasil population to define cognitive impairment (22). Next, participants with no diabetes at baseline who developed the disease during follow-up were classified as having diabetes in sensitivity analyses to investigate the impact of classifying these intermediate cases as having, as opposed to not having, diabetes on the incidence of cognitive impairment. We also analyzed risk among those with diabetes based on their level of control (glycated hemoglobin ≥8.0% vs. <8.0%) to verify the consistency of our findings. Next, we ran a competing-risk regression model considering death as a concurrent event (23). We also performed an additional analysis considering only TMT-B performance to identify cases of cognitive impairment and examine the influence of the familiarization step in the investigated association.

After excluding participants who died, were lost to follow-up, or had missing information on relevant variables, we analyzed data from 12,214 middle-aged and older adult participants (Supplementary Fig. 1).

Overall, a majority were women (n = 6,778 [55.5%]) and White (n = 6,400 [53.0%]) and had at least a college degree (n = 6,744 [55.2%]). Obesity was present in 22.3%, hypertension in 33.4%, and a common mental disorder in 26.6% of the participants. Only one in five participants was classified as physically active (23.5%), and 14% had diabetes at baseline. Stratifying the sample according to diabetes and physical activity status, a higher proportion of women were inactive among those with and without diabetes. Also, the group of inactive participants with diabetes had a higher proportion with lower educational attainment (elementary school or less), obesity, and hypertension (Table 1).

The mean follow-up time was 8.1 (SD 0.6) years. Regression models of global cognitive scores showed that having diabetes and being physically inactive predicted an accelerated yearly decline in global cognitive function. Participants with diabetes who were active at baseline had slower declines compared with those with diabetes who were inactive (Table 2). The specific trajectories of global cognition throughout adulthood are illustrated using age as a random effect and time scale (Fig. 1). The sum of the decrements in cognitive function associated with diabetes and with physical inactivity resulted in a lower cognitive function z score for those with both risk factors (orange line) at any age. These cognitive declines advanced with increasing age.

After excluding an additional 1,587 participants with cognitive impairment at baseline, we found that 885 (8.3%) developed cognitive impairment during follow-up. Having diabetes and being physically inactive were each associated with a higher risk of cognitive impairment, with a significant interaction term (diabetes × physical inactivity P = 0.072). Inactive participants had a greater risk of cognitive impairment regardless of their diabetes status; however, the risk was particularly high (hazard ratio [HR] 1.71; 95% CI 1.22, 2.39) when inactivity was combined with having diabetes; among those with diabetes who were physically active, we found only a small nonsignificantly increased risk (HR 1.18; 95% CI 0.73, 1.90) over our follow-up (Table 3). In a direct comparison of the two groups, physically inactive participants with diabetes showed a nonsignificantly higher risk of developing cognitive impairment compared with active participants with diabetes (HR 1.45; 95% CI 0.92, 2.29).

In Laplace regression, we found that the onset of cognitive impairment was earlier in inactive versus active adults. Compared with active adults without diabetes, inactive adults without diabetes experienced onset, on average, 3.13 (95% CI 2.04, 4.23) years earlier, and inactive adults with diabetes experienced onset 3.45 (95% CI 1.89, 5.00) years earlier. However, active adults with diabetes showed only a considerably smaller nonsignificant difference of 0.66 (95% CI 0.86, 2.18) years (Table 3). Among those with diabetes, being active was associated with a delay of 2.73 (95% CI 0.94, 4.51) years in the onset of cognitive impairment.

Our findings were only slightly changed when we alternatively applied education- and age-specific cutoffs to classify cognitive impairment (Supplementary Table 1). In these analyses, physical inactivity was associated with an increased risk of cognitive impairment in adults with (HR 1.34; 95% CI 1.08, 1.67) and without (HR 1.57; 95% CI 1.21, 2.02) diabetes. Diabetes continued to be not associated with increased risk of cognitive impairment in active participants (HR 1.23; 95% CI 0.81, 1.85). Results were also similar in additional analyses including the 1,456 incident cases of diabetes ascertained during follow-up (Supplementary Table 2). Analyses considering current diabetes control based on HbA_1c levels were also consistent with the main findings; those who were inactive had a greater risk of developing cognitive impairment independent of the level of glycemic control; however, physically active participants with poor control did not present an increased risk of incident impairment (HR 1.09; 95% CI 0.79, 1.49) (Supplementary Table 3). Competing-risk regression model results considering death as a concurrent event were also consistent with the main analyses; the risk of incident cognitive impairment was greater in inactive participants regardless of diabetes status (nondiabetes: HR 1.27; 95% CI 1.05, 1.53; diabetes: HR 1.50; 95% CI 1.19, 1.90). However, active participants with diabetes showed no increased risk compared with active individuals without diabetes (HR 1.21; 95% CI 0.85, 1.72) (Supplementary Table 4). Finally, a sensitivity analysis considering only the TMT-B score as a criterion for cognitive impairment confirmed the association between physical inactivity, with an increased risk of diabetes in participants with (HR 1.54; 95% CI 1.12, 2.13) and without diabetes (HR 1.20; 95% CI 1.00, 1.43) (Supplementary Table 5).

In this sample of middle-aged and older Brazilian adults participating in the ELSA-Brasil study, we found that both having diabetes and being physically inactive were associated with faster cognitive decline, higher risk of cognitive impairment, and earlier onset of cognitive impairment. However, being physically active attenuated the association with diabetes and, among those with diabetes, that with worse glycemic control in these outcomes. Additionally, being physically active was associated with delayed onset of cognitive impairment in those with diabetes.

Aging produced a major cognitive decline in all groups examined. The presence of diabetes increased this decline, making those with diabetes more susceptible to reaching the level of cognitive impairment. Among those with diabetes, the additional decline associated with physical inactivity, although small when assessed yearly, was large enough to produce, over time, a difference in cognitive level sufficient to result in both an earlier age at onset and an increased risk of cognitive impairment, both of a clinically relevant magnitude.

Previous studies have shown the interplay of physical activity with other risk factors in reducing the risk of cognitive decline and dementia (24). Findings from the UK Biobank indicate that air pollution may weaken the protective effects of objectively measured physical activity against dementia (25). Conversely, physical activity seemed to mitigate the association of cardiovascular risk factors (e.g., hypertension and obesity) and aging with incident dementia among middle-aged and older adults in England (8,9). Additionally, physical activity seemed to counteract the risks associated with sedentary behavior, reducing the overall risk of all-cause dementia, as suggested in the investigation of 90,320 middle-aged and older adults from the UK Biobank. Those with high levels of sedentary behavior and low levels of physical activity had an increased risk of all-cause dementia and Alzheimer disease, whereas those with high physical activity levels did not (7).

Although the literature highlights the importance of physical activity in the prevention of cognitive impairment among those with diabetes, only a few studies with relatively small numbers of participants have investigated the joint association of physical activity and diabetes with regard to cognitive outcomes. In the Swedish cohort cited earlier (13), adults with diabetes and low physical and mental activity (physical activity, social network, work complexity, and education) showed a higher risk of dementia compared with those who were diabetes free and engaged in moderate to high levels of such activity. The same study found similar brain volumes between adults with active lifestyles with and without diabetes, whereas less active patients with diabetes had the smallest brain volumes (13). Another cross-sectional study of 2,646 older Chinese adults with diabetes showed that physical activity was associated with lower likelihood of cognitive impairment, partially through improved sleep quality and decreased depressive symptoms (26). A previous systematic review with data from five randomized controlled trials (N = 738) revealed that exercise can improve cognitive function in adults with type 2 diabetes (11). However, the prevalence of physical inactivity remains elevated in individuals with diabetes (27). Therefore, our findings reinforce the need for public health policies promoting physical activity among those at high risk of cognitive impairment and dementia, including individuals with diabetes.

Diabetes is associated with impaired vascular health, which increases the risk of stroke, heart disease, and vascular dementia (28). Damaged blood vessels disrupt blood flow to the brain, worsening the delivery of nutrients. Also, impaired insulin action, a hallmark of type 2 diabetes, has been shown to increase the risk of multifocal ischemic lesions, which can lead to cognitive impairment and dementia (29). A study involving 5,586,048 participants age ≥40 years revealed that insulin resistance, as estimated by the triglyceride glucose index, was associated with an increased risk of incident dementia independent of diabetes status (30). In addition, hyperglycemia seems to play a role in the association between diabetes and dementia. High blood glucose level is related to a dramatically more toxic β-amyloid protein accumulation, one of the hallmarks of Alzheimer disease, in cells lining blood vessels in the brain (31). Physical activity is a valuable nonpharmacological approach to reducing the incidence and severity of diabetes and many of both its cardiovascular and microvascular complications.

In the Diabetes Control and Complications Trial and the subsequent follow-up evaluation, the Epidemiology of Diabetes Interventions and Complications study, elevated HbA_1c concentration was associated with impaired motor speed and psychomotor efficiency across an 18-year follow-up (32). The initial findings of the Action to Control Cardiovascular Risk in Diabetes (ACCoRD-mInD) study revealed an association between HbA_1c levels and cognitive function among individuals diagnosed with type 2 diabetes (33). Prolonged high blood glucose levels are linked to microvascular alterations, including retinopathy and nephropathy (34). Moreover, retinal microvascular abnormalities have been associated with cognitive decline and dementia, suggesting that these changes could serve as indicators of cerebral microvascular disease (34). In addition, poor glycemic control and diabetes duration were associated with an increased risk of conversion from mild cognitive impairment to dementia in middle-aged and older adults with diabetes (10). We identified that LTPA was particularly beneficial in those with HbA_1c ≥8%, reducing their excess risk from 53% to a nonsignificant 9%. Within this context, our findings provide novel evidence that physical activity is associated with improved cognitive outcomes in adults with diabetes independent of disease duration and control status, highlighting the importance of promoting physical activity among individuals with diabetes.

In light of the forecasted burden of dementia in Brazil and worldwide, the discovery of strategies capable of delaying the onset of cognitive impairment, a condition that usually precedes dementia, is critical. Here, we first showed that physical activity was linked to a slower rate of cognitive decline, to the point of observing a 3-year delay in the onset of cognitive impairment in adults living with diabetes. Delaying the onset of cognitive impairment seems to have broad economic implications. Globally, the economic cost of dementia reached U.S.$1.3 trillion in 2019, primarily as a result of unpaid informal care (1). However, a subtle decline in memory and executive function observed in cognitive impairment may have economic consequences, such as loss of productivity (35). For example, a Swedish study showed that an increase in the Clinical Dementia Rating Score from 0.5 (i.e., mild cognitive impairment) to 1.0 (i.e., mild Alzheimer disease) corresponded to U.S.$5,700 in excess costs, suggesting that delaying the transition to Alzheimer disease may result in considerable economic benefits (36). A novel strategy delaying the onset of Alzheimer disease by 3 years in individuals age >70 years would result in a decrease of ∼29% in prevalence and U.S.$500 billion in the cost of Alzheimer disease in 2050 in the United States (37). The anticipated increase in the number of individuals with dementia underscores the critical need for research focused on discovering disease-modifying treatments and effective low-cost interventions for preventing or delaying cognitive impairment onset.

In this study, we present new evidence indicating that engaging in physical activity can potentially lower the rate of cognitive decline and the risk of cognitive impairment in individuals with diabetes who are at risk of developing dementia. The escalating incidence of cognitive impairment and dementia has been closely associated with the high prevalence of cardiometabolic risk factors, including diabetes and physical inactivity. The combined burden of diabetes, physical inactivity, and dementia becomes evident as these conditions become more prevalent with advancing age. For instance, in Brazil, individuals age ≥65 years had a diabetes prevalence (21.6%) seven times greater than those age 35–44 years (3.1%) in 2019 (38). Moreover, physical inactivity is 2.1 times more prevalent in older adults (age ≥60 years) than in young adults (age 18–24 years) (39). A population-based study involving middle-aged and older adults in Brazil revealed that 15% of dementia cases could be attributed to physical inactivity and diabetes (4). In contrast, a survey conducted in 2019 showed that globally, two of three individuals believe that dementia is a normal part of aging, and one in four believe that nothing can be done to prevent it (14). Despite the World Health Organization’s Global Action Plan on Public Health Response to Dementia setting targets to reduce the prevalence of physical inactivity by 10% and control the rising incidence of diabetes by 2030, achieving these goals seems unlikely, particularly in Brazil (1). Our findings add to previous research recognizing the major role of physical activity in public health policies that safeguard cognitive function in individuals at high risk of dementia, such as adults living with diabetes.

Our study has some limitations to be highlighted. First, objectively measured physical activity may provide a more accurate measure of daily movement. However, findings on the investigated association have been similar in studies using questionnaires and devices (40). We did not investigate these associations using other domains of physical activity (e.g., household and occupational). Second, reverse causality may not be completely ruled out, even after excluding participants with cognitive impairment at baseline. Third, based on an observational study, we cannot rule out unmeasured or residual confounding. Finally, although we highlighted the heterogeneous reliability of the cognitive tests adopted in the study, our results were consistent when we used only TMT-B scores to identify cognitive impairment cases.

Our study also has strengths that merit mention. ELSA-Brasil is a large cohort of middle-aged and older Brazilians with highly standardized sociodemographic, aboratory, and cognitive measurements and comprehensive assessment of diabetes status. In addition to its multiethnic and LMIC context, to our knowledge, ours is the first large study of individuals with diabetes investigating the joint effects of physical activity and diabetes on incident cognitive impairment in a diverse sample in terms of age, sex, and educational level.

In conclusion, our findings reveal that LTPA may attenuate the rate of cognitive decline in middle-aged and older adults living with diabetes, potentially delaying the onset of cognitive impairment in this population. Although future studies, including randomized clinical trials, are warranted, our results strengthen the support for a pivotal role of physical activity in mitigating diabetes-related cognitive complications.

Funding. This study was partially funded by the Department of Science and Technology, Brazilian Ministry of Health (Ministério da Saúde), and the Brazilian Ministry of Science, Technology and Innovation (Financiadora de Estudos e Projetos and Conselho Nacional de Desenvolvimento Científico e Tecnológico [CNPq]), through grants 405545/2015-0 RS, 405551/2015- 0 BA, 405543/2015-8 ES, 405552/2015-7 MG, 405547/2015- 3 SP, and 405544/2015-4 RJ. N.F. was supported by a postdoctoral fellowship from CNPq (150161/2023-9).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. N.F. wrote the first draft of the manuscript. N.F., D.d.P., R.C.P.d.R., B.B.D., and M.I.S. were involved in the conception, design, and conduct of the study and the analysis and interpretation of the results. D.R., A.L.P., S.M.B., and C.K.S. provided intellectual contributions with critical inputs to improve manuscript quality, in addition to participating on the scientific board of the ELSA-Brasil study and in funding acquisition. All authors edited, reviewed, and approved the final version of the manuscript. N.F. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.